Track record

The dashboard we built to keep ourselves honest.

Every prediction every brief publishes is logged, resolved against actual market outcomes, and aggregated into the per-bucket calibration view below. Updated nightly.

Last refreshed: 7/6/2026, 3:12:09 am IST

Calibration window mix: 35% of the last 90 days’ resolved predictions were backfilled from historical bundles (retrospective replay), and 65% were collected live. We disclose this explicitly because retrospective replay can encode hindsight; treat the displayed numbers as a directional read of the engine’s calibration, not a live-only track record.

Drift flags

  • options_strike · very_high

    calibration_error +0.11 over 7d window (drift threshold 0.08)

Calibration error - last 90 trading days

Mean predicted probability minus actual hit rate, per prediction type. Dashed lines mark the ±8% drift tolerance.

Per-bucket calibration

A bucket is calibrated when its hit rate is close to its mean predicted probability - the calibration error column should hover around zero. Errors larger than |0.08| trigger a drift flag (above) and a note in the next brief’s confidence section.

touch watch

BucketnHit rateMean predictedCalibration error
very_high13684%86%+0.02
high17974%72%-0.02
moderate7362%57%-0.05
low24231%32%+0.01

avoidance

BucketnHit rateMean predictedCalibration error
very_high21990%86%-0.04
high17376%72%-0.04
moderate20955%57%+0.02
low20434%32%-0.02

options strike

BucketnHit rateMean predictedCalibration error
very_high22088%86%-0.02
high19269%72%+0.03
moderate10461%57%-0.04
low13229%32%+0.03

These numbers are computed from a sanitised public view of the outcome log. Specific prediction ids, symbols, and levels are never exposed on this surface. Subscribers see the full per- prediction detail in their portal.